Circuits, Devices, Networks, and Microelectronics CHAPTER 17. LINEAR ACTIVE FILTERS AND FREQUENCY PROFILES 17.1 INTRODUCTION All circuits have a frequency response, characterized by their connected set of reactive components and devices and their mix of intrinsic time constants. In many instances it is desired that an analysis of the frequency character of a circuit be accomplished. But in other instances it is desired to adjust the circuit’s frequency behavior and define or redefine the frequency profile of the circuit. In most instances a frequency profiles is defined in terms of amplitude ‘pass’ functions. There are four basic types: (1) (2) (3) (4) low-pass high-pass band-pass band-stop Unless semiconductor components dominate the circuit, its time constants should be expected to relate to capacitances and inductances and resistances. In this respect, intermediate frequency and RF (radiofrequency) profiles may be defined by pF or fF capacitances and H inductances. Profiles also may be defined by RLC resonance networks and the placement of poles and zeros. High RF profiles usually need to accommodate the effects of parasitic components such as leakage paths, wiring inductances, and fringing capacitances. Frequency profiles in the UHF range (300MHz to 3GHz) or higher will relate to microstrips, resonant cavities or artificial crystals for realizations of desired frequency responses. At lower frequencies, typically associated with audio systems or other biological interface systems, many components may be large and cumbersome. Consequently techniques have been developed in which amplification loops are used to define profiles. These circuit topologies are usually called active filters. Active filters are also used at higher frequencies in order to be able to deploy the function on an integrated circuit, where all components will be small in both value and. size An active filter is a frequency-responsive network driven by one or more active drivers. The typical driver typically is an opamp or one of its cousins. Often the circuit can then be created in an integratedcircuit form in which the only frequency-dependent components are capacitances. The integrated circuit active filter is also appropriate to RF (radio frequencies) as a means to either avoid or to accommodate the parasitic wiring capacitances and inductances. 459 Circuits, Devices, Networks, and Microelectronics A frequency profile is generically defined by a transfer function T(s) = |T(s)| ej(s) in the s-domain, for which s = j. And as identified by chapter 4, the graphical representations of |T(s)| and (s) are called Bode magnitude and Bode phase plots, respectively, after their originator, Hendrik Bode (1905-1982). The order of the profile T(s) is defined by the number of frequency-dependent components. All profiles will have a numerator N(s) and a denominator D(s) T ( s) N ( s) D( s ) (17.1-1) for which the numerator zeros are called zeros and the denominator zeros are called poles. Poles are associated with roll-off of the frequency profile. Zeros will have the effect of ‘roll-upward’. All transfer functions have the same number of zeros as they do poles. The effects of poles are almost always visible. The effects of zeros may be less visible since many zeros will fall either at zero or at infinity. A great deal of attention has been given to the art of frequency profiling, for which placement of poles and zeros is of critical importance to the character of the profile. Therefore many profile options will be identified in terms of names of their benefactor or in terms of mathematical descriptors. We will not attempt to survey all of the different types except to say that there are enough to satisfy most of the discriminating engineering requirements. Usually the different types will exist in tabulated forms. For simplicity and conservation of context the table values for profiles of interest can be assumed and then we will load and go, provided we know how to rescale and convert to the frequencies of interest and have enough perspective to make a judicious choice. Bilinear circuits: The bilinear circuits are the simplest category of frequency profile. The transfer function T(s) is a ratio of two linear functions, otherwise called a bilinear form. The general form for the bilinear function is T (s) K sz s p K' 1 s z 1 s p (17.1-2) Table 17.1-2 shows the magnitude response for the two single-time-constant types of bilinear response (low-pass and high-pass) and their accompanying phase responses. The roll-off slope is of the form of –20dB/decade (which is the same as a ratio of -1/1). The roll-off slope on the log-log plot is linear, since for for > 10p, the magnitude of the denominator becomes approximately linear, i.e. D( s) 1 p p 2 as >> p 460 Circuits, Devices, Networks, and Microelectronics Table 17.1-1(a). Bode magnitude and phase response for low-pass profile. Table 17.1-1(b). Bode magnitude and phase response for high-pass profile. Bode magnitude plots are usually represented as log-log scales with the frequency axis of the form of a decade scale and the amplitude axis of the form of a magnitude scale in dB. The zeros are usually at either infinity or zero (which is typical) and beyond the range of a finite Bode plot. For a more emphatic roll-off, higher-order transfer functions are required. Assuming appropriate placement of poles and zeros roll-off becomes as much as ~ n x (-20dB/dec), where n = number of poles, The case should be stated for more than the |T(s)| roll-off in terms of the low-pass profile (Table 24-1.1), Low-pass is an inevitable fact of all electronic circuits. The higher the frequency the more difficulty the signal has of passing through a network and the more readily it is shunted to ground. And so T (s ) 0 . Biquadratic circuits: The simplest categories of frequency profiles are (1) first-order forms and (2) second-order forms. The second-order are profiles are called biquadratic forms. The biquadratic form yields a complete set of pass-band profiles as represented by Table 17.1-2. The general form of the biquadratic function is T ( s) K n 2 s 2 n1 s n0 s 2 s 0 Q 02 K' 1 s a s2 b 1 s 0 Q s 2 02 (17.1-3) Table 17.1-2 shows the types of magnitude response for different numerator functions. For convenience and definition of the key profile parameters, the denominator function is always of the form 461 Circuits, Devices, Networks, and Microelectronics D(s) s 2 s 0 Q 02 D(s) 1 s 0 Q s 2 02 or (17.1-4) Equation (17.1-4) is the preferred form for the quadratic denominator since it relates to a characteristic (resonance) frequency 0 = 2f0 and a ‘quality factor’ Q. The band-pass function, for which n2 = 0 and n0 = 0, will have peak magnitude at f0 and will be symmetric about f0 on the decade frequency scale. Low-pass and high-pass quadratic functions will also show a profile peak if Q > 1 2 , although it will not be very pronounced until Q > 2. Type (Amplitude normalized) transfer function (a) Band-pass T (s) s 0 Q s s 0 Q 02 2 (Q = 4 represented) (b) Low-pass T (s) 02 s 2 s 0 Q 02 (Q = 4 represented) (c) High-pass T (s) s2 s 2 s 0 Q 02 (Q = 4 represented) (d) Band-stop T (s) s 2 s 0 Q ' 02 s 2 s 0 Q 02 (Q = 4 represented) 462 Magnitude Profile Circuits, Devices, Networks, and Microelectronics (e) All-pass T (s) s 2 s 0 Q 02 s 2 s 0 Q 02 (Q = 4 represented) Table 17.1-2. The equations shown are normalized to |T(max)| = 1.0 for the bandpass, |T(0)| = 1 for the low pass and |T(∞)| = 1 for the high-pass. Biquadratic profiles usually serve as benchmarks for their higher-order cousins. The profile for the profile 17.1-2(e) (all-pass) bears some explanation. It is more correctly identified as a phase-shift circuit for which the phase varies as a function of frequency but the amplitude does not. Since it is of second-order it will shift phase by = -2 x 90o = -180o across the span 0 < < ∞ . The normalized form of the band-pass function, for which n1 = 0/Q , is a means to identify the reason why Q is called the quality factor of the profile. Since the band-pass function is a means to select and pass a given frequency it is appropriate to identify the bandwidth of the profile = 2 – 1 , which is defined in terms of the 3-dB levels ( T 1 2 ), a.k.a. half-power levels as indicated by figure 17.1-3. Figure 17.1-3. (Normalized) biquadratic band-pass function Analysis of the normalized band-pass equation (given in Table 17.1-2(a) at magnitude levels for which T 1 2 , gives a quartic equation with four solutions: 0 2Q 0 2Q 1 4Q 2 (17.1-5) Of these solutions the only ones that are greater than zero are 463 Circuits, Devices, Networks, and Microelectronics 1 for which 0 2Q 0 2Q 1 4Q 2 2 1 and 2 0 2Q 0 2Q 1 4Q 2 0 (17.1-6a) Q which represents the selectivity characteristic of the resonance peak, i.e. Q 0 f0 f (17.1-6b) which is why Q is called the ‘quality factor.’ The mathematics is detailed in Chapter 4 and equation (17.1-6b) is the same as equation (4.5-11). Otherwise emphasis should be made that Q is a magnitude factor for the low-pass profile resonance peak since Q T ( 0 ) (17.1-7) T ( 0) and which was examined more carefully in chapter 16 as well as in chapter 4. As represented by figure 17.1-2 the low-pass and high-pass profiles of primary interest are (1) Q = 1 2, for which the profile is maximally flat and (2) Q > 4 , for which the resonance peak begins to dominate. Figure 17.1-2: Low-pass profile for several values of Q, i.e. Q = {0.5, 0.707, 2, 4}. The resonance peak is approximately at f0 for Q > 4. There is more to the story, however, when we extend the reach to specific applications. They are categorized according to their properties as follows: (1) Butterworth profiles (2) Chebyshev profiles = maximally flat in the pass-band = maximum roll off at the pass-band corner 464 Circuits, Devices, Networks, and Microelectronics (3) Bessel (Thomson) = maximum linearity in the phase response and minimum overshoot for impulse signals There are others which may accept ripple in both pass-band and stop-band in order to achieve better quality response for one of the above categories. But each of the three categories cited above the biquadratic response may be related to Q as follows: (1) Butterworth: Q=1 (2) Chebyshev: (3) Bessel: Q = 1.0 Q = 0.5 2 (3dB ripple ) The Chebyshev profile is a little more artistic than suggested by these simple relationships, since it may be either of three types. Type (1) has ripple in the pass-band, type (2) has ripple in the stop-band and type (3) (a.k.a. elliptic or Cauer response) has ripple in both pass-band and stop-band. The Chebychev name is a consequence of the mathematics, for which the poles and zeros are defined by Chebyshev polynomials. In the s-plane its poles will pattern in the form of an ellipse. The ripple for the Chebyshev profile may also be of different relative magnitudes. As might be expected for the biquadratic form, a Q > 1.0 is also a citation for more ripple in the pass band and the benefit of a greater roll-off at the break point. The circuit designer then takes up a role as an artist in mathematics and circuit electronics. 17.2 NORMALIZED FORMS and FREQUENCY RESCALING Many circuit topologies have been analyzed, addressed, profiled and tabulated by generations of electrical engineering slaves. Tabulated forms are normalized to a characteristic frequency 0 =1.0 r/s or to a feature frequency 1 =1.0 r/s. The normalized form of the topology is typically the one to which rescaling techniques of different modality are applied. Frequency rescaling is the analytical process of rescaling the components of a circuit topology to one that will yield an exact copy of the profile at another frequency. Frequency rescaling can also be used to transform the circuit into an equivalent topology with the same profile, in which case it is called a frequency transformation. Circuit profiles is defined in terms of a set of zeros in the numerator N(s) terms and another set of zeros in the denominator D(s) defined as poles because of their effect on the profile. Poles and zeros will each have a relationship to a time constant. And if the time constants are rescaled to a new s’ =j’ plane then they will have the exact same profile if x1 = ' x2 . Frequency response is defined by time constants (duh!). For the simplest case, for which all of the frequency dependent components in a circuit are capacitances then 1 = R1C1 and 2 = R2C2, and the s and s’ planes will relate to one another by 465 Circuits, Devices, Networks, and Microelectronics R1C1 ' R2 C2 (17.2-1) where we are identifying that s = j and s’ = j’. So if we do not change resistance, i.e. R1 = R2, then (17.2-1) reduces to C1 ' C 2 (17.2-2a) and we can shift the profile from the s to the s’ plane by merely rescaling the capacitance according to C2 C1 ' (17.2-2b) For example if the s’ domain is a rescale of 1000 s, for which ’ = 1000, then the capacitances must be 1000 times smaller. A 1.0mF capacitance would become a 1.0F capacitance. If we do not change capacitance, i.e. C1 = C2, then R1 ' R2 (17.2-3a) and we can shift the profile from the s to the s’ plane by rescaling the resistance according to R2 R1 ' (17.2-3b) For example if the s’ domain is 1000s, (for which ’ = 1000, the resistances must be 1000 times smaller. (And for ’ = 1000 the circuit is 1000 times faster since RC is 1000 times smaller). Actually we would most likely change both R and C, in a process called RC rescaling. This process is accomplished by a two-step procedure. (1) We keep R unchanged, i.e. R1 = R2, and rescale the capacitances to a new frequency M, i.e. C1 M C2 for which and for which M C1 C2 (17.2-4a) R1C1 M R1C2 For step (2) we hold C2 constant and rescale the resistances, for which 466 Circuits, Devices, Networks, and Microelectronics M R1C2 ' R3C2 R3 for which M R1 ' (17.2-4b) Now both R and C are changed and the profile is translated into the s’ plane to an identical profile which usually is at a higher frequency. The process in a different order, and is indicated by figure 17.2-1. The most likely option is to elect a final capacitance value C2 and then let the resistance follow. Equations (17.2-4a) and (17.2-4b) represent the conventional RC rescaling algorithm: M and Cinit init C final R final (17.2-5a) M Rinit final (17.2-5b) The frequency M in equations (17.2-5) is defined as the rescaling frequency. The init and final are initial and final frequencies, respectively. The mathematics is further simplified if the initial frequency init = 1.0 r/s, i.e. that associated with a normalized profile. The same process can be accomplished with inductances except by means of the time constant L / R for which the equivalent to equation (17.2-1) is then L1 R1 ' L2 R2 (17.2-6) but it is not common to rescale according to inductances. A diagrammatic representation of RC rescaling using equations (17.2-5) is shown by figure 17.2-1. 467 Circuits, Devices, Networks, and Microelectronics Figure 17.2-1 RC rescaling steps as represented by two rescaling paths. The upper path is usually the path of choice and matches equations (17.2-5). There are several good candidates for RC rescaling that we will check out in later sections of this chapter. For topologies that include both capacitances and inductances the rescaling process is not quite as simple. For example, the RLC biquadratic circuits will have 02 1 LC R 1 1 1 L RC L C (17.2-8) So the resistances cannot be rescaled by the simple time-constant method applied to L and C separately. A different paths must be followed which includes an impedance magnitude rescaling. The rescaling algorithm will then be of the form R2 k M R1 C2 C1 k M k f L2 L1 k M k f (17.2-9(a) (17.2-9(b) (17.2-9(c) Quadratic rescaling must then be accomplished by means of either (1) a component translation which retains the simplicity of equations (17.2-1) or (17.2-7) or (2) paths like those represented by equations (17.2-9). Since the translation is essentially an equivalent circuit form it invites usage of amplifier-driven components and the reconfiguration of the topologies into an active-filter form. 468 Circuits, Devices, Networks, and Microelectronics 17.3 BIQUADRATIC CIRCUIT TOPOLOGIES There are a number of RC topologies that can be invited to the biquadratic table. Almost all of them employ one or more gain elements, typically of the form of an ideal opamp, which certainly has its caveats if it turns real. But otherwise the opamp is a reasonable drive element for applications in which biquad profiles (table 17.1-2) are elected. For example the Sallen-Key circuit topology shown by figure 17.3-1 is a benchmark standard. It is simple, requires only one amplifier component, and the Q of the circuit is separately tunable relative to the characteristic frequency 0. Figure 17.3-1 Sallen-Key single-amplifier biquad. Nodal analysis at v1 and v+ gives v1 (G1 G2 sC1 ) vo sC1 v s G1 v G2 0 v (G2 sC 2 ) v1G2 0 Using vo Kv , where K 1 Rb Ra gives T ( s) G G vo K 1 2 vs C1C 2 2 G2 1 K ) G1 G2 s s C2 C1 G1G2 C1C 2 (17.3-1) Typically we let C1 = C2 = C and G1 = G2 = G for which T ( s) v0 K 02 2 v s s s 0 3 K 02 (17.3-2) where 0 G C 1 / RC . Note that (17.3-2) is of the low-pass form, with Q = 1/(3 – K). So Q can be tuned by means of an adjustment of the ration of Rb/Ra according to 469 Circuits, Devices, Networks, and Microelectronics Q 1 2 RB / R A (17.3-3) Since the ratio G2/C2 is consistent throughout equation (17.3-1) we may taper the Sallen-Key biquad by selecting C2 = aC1 = aC and G2 = aG1 = aG . This modification does not change the characteristic frequency 0, but will change the form of the expression for quality factor to the form Q 1 1 a Rb / Ra (17.3-4) In particular, if the gain element is set up for K = 2 (for which Rb/Ra = 1) as shown by figure 17.3.2 and Q is defined by tapering factor a, then the circuit simplifies to Q 1 a Figure 17.3-2. Design-B of the Sallen-Key for tunable Q. Note that we have set the design in normalized form (0 = 1.0r/s), to which we can apply the RC rescaling algorithm (17.2-5) as shown by figure 17.3-3. In this realization you might note that the parameter rescaling forms have made usage of the approximation 1 2 0.16 . Figure 17.3-3. Design B of the Sallen-Key with RC-rescaling to10kHz and Q stepped through several values. Also take note that the zero-frequency gain K = 2.0. Another form of the Sallen-Key topology is the Saraga design shown by figure 17.3-4. 470 Circuits, Devices, Networks, and Microelectronics Figure 17.3-4. Saraga low-sensitivity design for the Sallen-Key single-amplifier biquad with 0 normalized to 1.0 r/s. Note that for the Saraga design, R1 = 1/Q. The topology may then be rescaled to the frequency domain of interest by RC rescaling. We can use any number of active components in a circuit, and sometimes it is of practical benefit to do so as represented by the two-integrator loop of figure 17.3-5. It has three output points, each of which differs by the factor –0/s. Figure 17.3-5. Two-integrator loop. The factor –0/s is generated by the ‘Miller’ integrator which is a straightforward and simple circuit. The two-integrator loop consists of two Miller integators and one inverter-summing circuit as shown by figure 17.3-5, for which the output of the summing circuit will be vo 2 1 0 vo 20 vo n2 v I Q s s (17.3-5) 471 Circuits, Devices, Networks, and Microelectronics Collecting like terms and resolving the transfer function from vI to vo yields high-pass transfer function THP ( s) vo n2 s 2 2 v I s s 0 / Q 02 (17.3-6) From this equation and the relationship between stages n 2 0 s v1 vo 0 2 vI vI s s s 0 / Q 02 TBP (s) (17.3-7a) band-pass n2 02 v 2 vo 0 2 v I v s s s 0 / Q 02 TLP (s) (17.3-7b) low-pass and so two-integrator loops are also called state-variable filters since they will provide the three basic functions of low-pass, high-pass, and band-pass. The two-integrator loop most often deployed is the Tow-Thomas, or ring-of-three topology shown by figure 17.3-6. One of the integrators is a lossy integrator, and it is this choice that gives the circuit its operational benefit. Figure 17.3-6. Tow-Thomas (ring-of-three) two integrator topology. Nodal analysis at the input of opamp U1 gives 0 v s G3 v2 G4 v1 G1 sC1 (17.3-8a) and since U3 is a unity-gain inverter and U2 is a Miller integrator then 472 Circuits, Devices, Networks, and Microelectronics v2 G2 v1 sC 2 (17.3-8b) Solving for v1 we get s G3 C1 v1 2 v s s s G1 C1 G2 G4 C1C 2 TBP (s) (17.3-9) band-pass TLP (s) (17.3-10) low-pass and from equation (17.3-8b) we get G2 G3 C1C 2 v2 2 v s s s G1 C1 G2 G4 C1C 2 Typically we would (1) choose C1 = C2 = C and R2 = R4 = R, for which 0 1 / RC . Then (2) choose Q = R1/R. Then (3) choose amplitude k = R1/R3. Using this tuning algorithm, the profile parameters 0, Q, and k are separately adjusted in the order (1) thru (3). The tuning simplicity of the Tow-Thomas topology is even more evident when it is put in normalized form, for which we would let C1 = C2 = 1.0F and R2 = R4 = 1.0.Then R1 = Q and R3 = Q/k, and the topology is of the form shown by figure 17.3-7, which also includes an extra summing amplifier for the generation of the band-stop function. Figure 17.3-7. Tow-Thomas topology in normalized form configured with extra summing amplifier U4 for the generation of band-stop function. In normalized form equations (17.3-9) and (17.3-10) will be 473 Circuits, Devices, Networks, and Microelectronics v1 sk / Q 2 vs s s / Q 1 TBP (s) (17.3-11a) band-pass v2 k /Q vs s 2 s / Q 1 TLP (s) (17.3-11b) low-pass The band-stop output is provided by summing opamp U4 for which we have an extra output node v4 = vs + v2 and for which v4 vs v2 vs vs vs 1 sk / Q 2 s s / Q 1 s 2 s(1 k ) / Q 1 s2 s / Q 1 (17.3-12) Equation 17.3-12 is of the form of a band-stop function. Its behavior for a simulation in which k = 0.4 and Q = 4 is shown by figure 17.3-8. If we choose values such that k = 1, then the band-stop function becomes a band-reject form. Figure 17.3-8. Tow-Thomas simulation using tuning algorithm and normalized form as rescaled to 1.0kHz. Low-pass and band-stop outputs are shown. Peak amplitude of the low-pass is 0.4 and the minimum value of the band-stop is 0.6. Note that the Tow-Thomas topology acts as a filter and not as a resonator (i.e. no outputs have greater magnitude than vs). If we choose k = 2, then a special case biquadratic function occurs at v4 with transfer function T (s) s2 s / Q 1 s2 s / Q 1 (17.3-13) 474 Circuits, Devices, Networks, and Microelectronics This function is of the form of an all-pass function and serves only to produce a phase shift. The phase shift = -90o at the characteristic frequency f0 = 20. 17.4 BIQUADRATIC FILTER FUNCTIONS If we factor the biquadratic denominator D(s) s 2 s 0 Q 02 the roots are s 0 1 j 4Q 2 1 2Q and s 0 1 j 4Q 2 1 2Q (17.4-1) for which it is evident that the roots are complex when Q > 0.5. The s-plane is complex, and these roots are located in the left (negative) half of the s-plane. As the parameters 0 and Q, are varied a root-locus plot of these roots will result, as represented by figure 17.11. Figure 17.4-1. Root-locus plot for the poles of a biquadratic function. This figure is a modified copy of Figure 16.5-1 with more emphasis on the role of Q. In particular, we can see that the root-locus plot for 0 = constant is a circle with radius 0. If we determine the magnitude of either of the roots identified by equation 17.4-1, we find that s 0 2Q 1 4Q 2 1 and that different values of Q and for which 0 (17.4-2a) correspond to different values of the angle , as shown by figure 17.4-1, 475 Circuits, Devices, Networks, and Microelectronics cos 0 (17.4-2b) 2Q We often use the roots and root loci to identify different profiles according to location of the poles and zeros. For example low-pass Butterworth profile with a roll-off of n x (-20dB/decade), is of the form H (s) 2 and will have roots at k (17.4-3a) 1 s / j 0 2n s 0 e j ( 2k ) / 2n 0 e jk (17.4-3b) where k = 0, 1, 2 … n, and for which the equation looks like the same form of entertainment you used to have with the mathematics of complex variables. If n = 2, then the two roots that fall in the left half plane are at k = (3/4) and (5/4), both of which relate to the angle = 45o, for which Q = 1 2 . So the maximally flat profile is the Butterworth profile by another name. If n = 4, then the four roots that fall in the left half plane are at k = (5/8), (7/8), (9/8), (11/8). These correspond to two values of Q = 1/(2cos(22.5o)) = 0.541, and Q = 1/(2cos(67.5o)) =1.307, and can be accomplished by a couple of Sallen-Key topologies in cascade, as shown by figure 17.4-2. Figure 17.4-2 Use of Sallen-key topologies in cascade to generate 4th-order (maximally-flat) Butterworth response. Take note that the roll-off slope is –80dB/decade = same as 4 x (–20dB/dec). The root-locus plots are also of advantage to the use special mathematical functions to identify pole and zero distributions that will effect a sharper cut-off at the break frequency fc of the profile. A fairly popular class of sharp cutoff functions is based on Chebyshev polynomials according to transfer function 476 Circuits, Devices, Networks, and Microelectronics H ( s) k 1 cos C n ( / c ) 2 2 (17.4-4) 2 where the function C n ( x) n cos1 / c is a Chebyshev polynomial, and is probably familiar only to them who is well-versed in special mathematical functions. The factor is called the ripple factor and defines the ripple in the pass-band. As represented by figure 17.4-3, which shows a 3dB ripple, the ripple factor = 1. Cutoff frequency is referenced to f c 2 c . Figure 17.4-3. 4th-order Chebyshev profile with = 1. And if we indulge in the mathematics we would find that the choice of 01 = 0.951c , Q1 = 5.59 and 02 = 0.443c , Q2 = 1.075 for two Sallen-Key biquads in series will result in a -3dB Chebyshev 4th-order profile as shown by figure 17.4-4. Figure 17.4-3. Sallen-Key implementation of 4th-order Chebyshev profile with = 1. The accompanying trace is that for the 4th-order Butterworth profile. A source of biquadratic topologies with names and provenance is located under the Texas Instruments Split-supply Analog Filter expert. It not only offers topologies for the standard biquadratic forms but deploys them according to the more common filter properties (i.e. Butterworth, Bessel and Chebyshev). 477 Circuits, Devices, Networks, and Microelectronics 17.5 DOUBLY-TERMINATED RLC LADDERS and TRANSFORMATIONS The section on biquadratic forms is a small slice of the collection of frequency profiles that are standardized. For applications and realizations beyond the biquadratic there are two means by which we may glean higher-order profiles: (1) poles and zeros defined by polynomials (e.g. Butterworth, Chebyshev, Bessel) and (2) RLC tables. The polynomials are probably more mathematically satisfying but the tables are more convenient. It is not uncommon to define the rescaling and transformation algorithms in terms of a doubly-terminated RLC topology, such as is represented by figure 17.5-1. A subset of the tables are shown by figure 17.5-2, and credit for them belongs to one of signature resources for analog filter design (M.E.Van Valkenburgh, Analog Filter Design, HRW, 1982. Unfortunately much of its content is not in softcopy and so the tables have to be included in hardcopy. Figure 17.5-1. Doubly-terminated RLC ladder topologies (equivalent forms) 478 Circuits, Devices, Networks, and Microelectronics Figure 17.5-2. Doubly-terminated RLC ladder tables (this is only a subset). The strip index at the top beginning with L1 is associated with the RLC ladder above the tables and the strip index at the bottom beginning with C1 is associated with the RLC ladder below the tables. Direct simulations using the RLC ladder topologies and table values will yield a normalized low-pass profile normalized. The categories (or polynomial types) are listed above the tables. All of the values are in Ohms, Farads, or Henrys, depending on the component. The rest of the story is that the tables are the touchstone for a profile of any of the generic types, low-pass, high-pass, band-pass, band-stop. Profiles are realized by means of (1) transformation and (2) frequency rescaling. The mathematics of making a transformation is entertaining, usually with the idea that a transfer function T(s) is a function of impedance/admittance ratios and the fact that if we transform them in the same way, the ratio will remain the same. Add a translation to this context and a low-pass can be translated into an 479 Circuits, Devices, Networks, and Microelectronics equivalent form with two shoulders instead of one, if we regard the low-pass as a one-shoulder profile centered about zero. All that is needed is to shift the zero and the other shoulder emerges. But the greater benefit of the process is that an RLC circuit can be translated into a form that employs impedance-converter circuits with opamps to eliminate the bulky inductances. And that takes the process into a form that can be micro-miniaturized since a VLSI chip can literally contain dozens of opamps. The active-filter context (using opamps) is then resolved as an R-C circuit, which can be translated to whatever frequency characteristic is needed using the algorithm of figure 17.2-1 or its equations (17.2-5). And it surely makes life more simple than the mathematical divinations. For example take a look at a transformation called the RLC:CRD transformation, so-called because the R is recast as a C, the L is recast as an R and the C is recast as an impedance-conversion construct element (= D) called a frequency-dependent negative resistance (FDNR). The concept is illustrated by Figure 17.5-3. Figure 17.5-3. RLC:CRD transformation The technique is also called the Bruton transformation and is primarily a mathematical manipulation, as represented by equation (17.5-2) Figure 17.5-3(a) has transfer function T ( s) v2 1 sC v1 R sL 1 sC (17.5-1) And figure 17.5-3(b) has transfer function T ( s) v2 1 s 2C v1 R s L 1 s 2 C 1 s2D 1 sC ' R ' 1 s 2 D (17.5-2) Where (17.5-2) is (17.5-1) with both numerator and denominator multiplied by (1/s). And the change of labeling does not bother the computation. But the new component is an opamp construct derived from the Fleige topology, as represented by figure 17.5-3. 480 Circuits, Devices, Networks, and Microelectronics Figure 17.5-3. (a) Generalized impedance converter (GIC) topology and (b) translation into FDNR (= D component). Analysis of figure 17.5-3(a) is a matter of nodal analysis at each of the three nodes labeled v0. They are all virtually connected because of the nullator action of the opamps. Eliminating v1 and v2 from the resulting three equations gives Y1Y3Y5 Y2Y4Y6 0 (17.5-2) Which can be solved for Yin = -Y1 as Yin Y2Y4Y6 Y3Y5 (17.5-3) With the component choices indicted by figure 17.5-3(b) and making the choice R3 = R4 we have the desired FDNR behavior Yin s 2 C2 C6 R5 s 2 D (17.5-4) where the magnitude of D is then C 2 C6 R5 . The process is just a bit of mathematics. But the mathematics more easily accomplished if developed on a simulation platform as represented in previous section of this chapter. The process is illustrated by figures 17.5-4 and 17.5-5: 481 Circuits, Devices, Networks, and Microelectronics Figure 17.5-4(a) Doubly-terminated RLC topology for a 5th-order Chebyshev with 1.0dB ripple in the pass-band. Note that the parameter values are the number values from Chebyshev (C) of figure 17.5-2. Figure 17.5-4(b) Simulation results. Note that the breakpoint is at f0 = (1/2) x 1.0 rad/sec = 160 mHz. So this profile is the normalized form. Figure 17.5-5 Same as figure 17.5-4(a) except implemented in CRD form. The L’s are replaced by R’s, the R’s are replaced by C’s, and the capacitances are replaced by FDNR’s. The parameterization in figure 17.5-5(a) is doing all of the mathematics grunt work, to include the frequency rescaling. If you look at the very bottom of the figure you will see that the capacitance has 482 Circuits, Devices, Networks, and Microelectronics been chosen to be .01 F. And then the rescaling algorithm according to equations (17.2-5) is located in the parameter list in the lower left corner. Notice that the 1.0r/s normalized frequency is translated into 0.16 since the simulation is always in Hz. The units are omitted since it is the mathematics that carries the process along. The scaling factor shown as kR is applied to all resistances, exactly as prescribed by equation (17.2-5(b)). The diagram is a little busy, but you should take note of a few simplifications that are used and are not uncommon. Due to the RLC:CRD transformation the capacitances relate to the R = 1.0 termination resistances, and hence are normalized to 1.0F. The other capacitances, which belong to the FDNR, are dealer’s choice. So they might as well be elected to be the same, i.e. 1.0F capacitances. This choice also simplifies the (FDNR) value for D, which by equation (17.5-4) gives D = C2 C6 R5 = 1.0 x 1.0 x R5 = R5 (17.5-5) So it is not uncommon to deploy the CRD representation in the form of an equal capacitance realization. And that lets the rescaling be a relatively simple RC rescaling process, as indicated by the parametric usage in the lower right corner of figure 17.5-5(a). For the final frequency (labeled as fC in the figure), the rescaled CRD realization gives the frequency profile result as shown by figure 17.5-5(b). Figure 17.5-5(b) Simulation results. Both normalized profile and rescaled profile at fC = 10 kHz are shown as a comparison to confirm that the transformation/rescaling will realize the same profile but at a different frequency. 483 Circuits, Devices, Networks, and Microelectronics There are a few extra wrinkles that are necessary which the mathematics does not represent. The capacitances that replaced the terminations (at each end) leave the circuit components floating relative to ground, and therefore must be accompanied by leakage resistances (shown as 999M values). If this is not done the circuit simulator willabort and give an error this effect. The values of the leakage resistances are arbitrary, but large. The RLC-CRD process is a generic and common choice in active-filter design. Others of note are ‘LeapFrog’ design, which use a clever process to transform the RLC ladder to an equivalent form that will accomplish to a ‘low-pass to band-pass’ (LP:BP) transformation. But in the interest of keeping the topic area of frequency profiling reasonably compact, LeapFrog design will be omitted except for name acknowledgement. Regrets. 484 Circuits, Devices, Networks, and Microelectronics PORTFOLIO and SUMMARY RC frequency rescaling M Cinit init C final R final and M Rinit final Sallen-Key biquad (lowpass ) = tuneable Q 0 = 1/RC Q 1 1 a R f Rx where a = tapering factor Tow-Thomas biquad = biquad states for (1) low-pass, (2) band-pass, (3) band-stop, (4) all-pass 0 = 1/RC = 1.0 r/s normalized Normalized values: C1 = C2 = 1.0F R2 = R4 = 1.0 R1 = Q R3 = Q/k TBS (s) v S v1 v4 s 2 k / Q 1 vs vs s 2 s / Q 1 TBP (s) v1 sk / Q 2 vs s s / Q 1 TLP (s) v2 k /Q 2 vs s s / Q 1 TAP(s) = TBS(s) when k = 1 Butterworth profiles = maximally flat in the pass-band 485 Circuits, Devices, Networks, and Microelectronics Chebyshev profiles = maximum rolloff at the pass-band corner Bessel (Thomson) = maximum linearity in the phase response, minimum overshoot for impulse signals, minimum delay of signals Simulations: 5th-order low-pass. Profiles and pulse response Chebyshev (1.0dB) Butterworth Bessel 486